2011 Poster Sessions : Database Join Acceleration with FPGAs

Student Name : Jared Casper
Advisor : Oyekunle Olukotun
Research Areas: Computer Systems
In databases, join is a critical but expensive operation, making it a prime candidate for hardware acceleration. As systems with more and more memory become common, in-memory database operations are becoming more prevalent. Additionally, databases are being used in applications where the queries are not known before hand, making it impracticle to generate indices to accelerate the joins.

Traditionally the hash-join algorithm has been used for database joins; however, recent research from Oracle and Intel suggests that the sort-merge algorithm can do better on newer processors with wider SIMD units and less per-core bandwidth. This work leads to question what is the limit of join speed with effectively unlimited data-level parallelism and large amounts of dedicated bandwidth, both of which can be achieved using an FPGA.

Jared Casper is a Ph.D. Candidate working with Kunle Olukotun in the Pervasive Parallelism Lab in the Department of Computer Science. His research focuses on developing computer architectures that make it easier for the programmer to make effective use of the vast amount of resources now available in a modern heterogenous multi-core system. He received is Bachelor's degree in Electrical Engineering and Computer Science from MIT in 2005.